Skip to content

The Particle-In-Cell (PIC) Method

Intuition

Direct N-body costs \(\mathcal{O}(N^2)\) because every particle talks to every other. PIC's trick: particles never talk to each other directly — they talk to a grid. Deposit charge onto mesh points, solve one field equation on the mesh, interpolate the field back. Cost collapses to \(\mathcal{O}(N + M)\), and the physics gains something profound: the simulation is self-consistent — the fields that push the particles are the fields the particles made.

The PIC cycle

Every time step walks the same loop:

  1. Deposit — assign particle charge to mesh: density \(n_j\)
  2. Solve — Poisson equation on the mesh: \(\dfrac{d^2\phi}{dx^2} = n(x) - n_0\) (electrons against a uniform ion background)
  3. Differentiate — \(E_j = -\dfrac{\phi_{j+1} - \phi_{j-1}}{2\Delta x}\)
  4. Interpolate — field from mesh to each particle position: \(E(r_i)\)
  5. Push — advance particles with leapfrog: \(\dot r_i = v_i\), \(\dot v_i = -E(r_i)\) (normalized electrons)

Cloud-in-cell weighting

The deposit and interpolation both use linear (CIC) weights. A particle at \(r_i\) between grid points \(x_j\) and \(x_{j+1}\):

\[n_j \mathrel{+}= \frac{x_{j+1} - r_i}{\Delta x}, \qquad n_{j+1} \mathrel{+}= \frac{r_i - x_j}{\Delta x}\]
\[E(r_i) = \frac{x_{j+1} - r_i}{\Delta x}E_j + \frac{r_i - x_j}{\Delta x}E_{j+1}\]

Using the same weights both ways is not cosmetic: it guarantees momentum conservation (no particle self-force). Implementation: np.bincount for deposit, fancy indexing for interpolation.

The field solve

The 1-D periodic Poisson problem is a tridiagonal(+corners) sparse system solved directly each step with scipy.sparse.linalg.spsolve — exact, fast, and reused thousands of times.

What PIC actually simulates

Each computational particle is a super-particle standing in for millions of real electrons; the method samples the phase-space distribution \(f(x, v, t)\) with markers. PIC is thus a Monte-Carlo-flavored solver for the Vlasov equation — kinetic physics (Landau damping, beam instabilities, phase-space vortices) at a fraction of a full 6-D grid's cost. The price: statistical noise \(\propto 1/\sqrt{N_\text{per cell}}\).

Common mistakes

  • Mismatched deposit/interpolation schemes → self-forces, artificial heating.
  • Under-resolving the Debye length / plasma frequency. Rules of thumb: \(\Delta x \lesssim \lambda_D\), \(\omega_{pe}\Delta t \lesssim 0.2\) — else numerical grid heating corrupts everything.
  • Too few particles per cell. Noise masquerades as physics; growth-rate measurements need enough markers for the signal to clear the noise floor.

Knowledge graph position

Prerequisites: Leapfrog, Poisson solvers, N-body. Leads to: Two-stream instability, production plasma simulation.

Quiz

Q1 (conceptual). Why does PIC scale so much better than direct summation?

Answer

Particle–particle interactions are mediated by the mesh: deposit is \(\mathcal{O}(N)\), the 1-D field solve \(\mathcal{O}(M)\), interpolation \(\mathcal{O}(N)\) — versus \(\mathcal{O}(N^2)\) pairwise forces. The grid acts as a common bulletin board.

Q2 (computational). A particle sits exactly midway between two grid points. What CIC weights does it give each?

Answer

½ and ½ — linear weighting splits it evenly.

Q3 (MCQ). The self-consistency of PIC means:

  • (a) results are independent of \(\Delta t\)
  • (b) the fields are recomputed from the particles every step
  • (c) particles never cross cell boundaries
  • (d) energy is exactly conserved
Answer

(b). Field ← particles ← field, every cycle — the defining loop of plasma physics, closed numerically.